mirror of
https://github.com/garrytan/gstack.git
synced 2026-09-16 01:45:29 +02:00
v1.87.0.0 feat: add verified CSO audits and replayable repair bundles (#2852)
* feat(cso): add verified audits and replayable repair bundles * fix(cso): harden qualification and setup boundaries * fix(cso): assemble security canaries at runtime * fix(cso): bound release proof and maintenance work Co-Authored-By: OpenAI Codex <noreply@openai.com> * fix(cso): require complete evaluation reports Co-Authored-By: OpenAI Codex <noreply@openai.com> * fix(cso): replay expired snapshots from supplied source Co-Authored-By: OpenAI Codex <noreply@openai.com> * test(cso): synchronize DNS cancellation assertion Co-Authored-By: OpenAI Codex <noreply@openai.com> * chore(ship): exempt repository owner from liveness proof Co-Authored-By: OpenAI Codex <noreply@openai.com> * test(cso): make recheck retention overlap deterministic Co-Authored-By: OpenAI Codex <noreply@openai.com> * chore: bump version and changelog (v1.85.0.0) Co-Authored-By: OpenAI Codex <noreply@openai.com> * fix(cso): pass native release gates Co-Authored-By: OpenAI Codex <noreply@openai.com> * chore: move release to v1.86.0.0 Co-Authored-By: OpenAI Codex <noreply@openai.com> * fix(cso): resolve rechecks by finding Co-Authored-By: OpenAI Codex <noreply@openai.com> * chore: move release to v1.87.0.0 Co-Authored-By: OpenAI Codex <noreply@openai.com> * fix(cso): pass macOS and Windows release gates Normalize BSD wc output, compare Windows paths by filesystem identity, preserve portable snapshot race coverage, and narrow POSIX-only Windows fixtures. Co-Authored-By: OpenAI Codex <noreply@openai.com> * fix(cso): harden native verification gates * fix(cso): refine Windows native diagnostics * test(cso): isolate Windows Git startup failure * test(cso): stabilize Windows native diagnostics * fix(cso): support hardened Git on Windows * fix(cso): close final verification gaps * test(cso): bound cold Docker fixture setup * fix(cso): restore cross-platform free-suite gates --------- Co-authored-by: OpenAI Codex <noreply@openai.com>
This commit is contained in:
Garry Tan
co-authored by
OpenAI Codex
OpenAI Codex
parent
9f81911136
commit
4a3c6a8a3c
@@ -0,0 +1,194 @@
|
||||
# Building and qualifying CSO runtimes
|
||||
|
||||
`bun run build:cso` compiles the trusted launcher, helper, and independent
|
||||
watchdog. Bun must support all four `--no-compile-autoload-*` switches. macOS and
|
||||
Linux also require a C compiler (`cc`, Clang, or GCC). Linux links the launcher
|
||||
statically; macOS signs it with the hardened runtime. `bun run build` includes
|
||||
this step and remains a strict release/developer build. Setup feature-probes the
|
||||
four Bun switches and the host native toolchain before building. If either is
|
||||
unavailable, it builds the rest of gstack, removes any stale CSO helpers, and
|
||||
reports `/cso` unavailable; the skill then reports `not assessed` and never
|
||||
substitutes repository tooling. Re-run setup after installing the prerequisite.
|
||||
When the probe succeeds, setup keeps the strict failure behavior and checks the
|
||||
artifacts for freshness. The generated
|
||||
`bin/gstack-cso-launcher` is the setup-built direct command; the compiled core
|
||||
is not the entrypoint. The core's direct-invocation marker catches accidental
|
||||
misuse; it does not authenticate a process against the trusted same-user host.
|
||||
The native launcher is the pre-runtime boundary that strips Bun, Node, and
|
||||
loader injection variables. CSO is not exposed as an npm package bin: its native
|
||||
launcher, compiled core, watchdog, and hidden `.gstack-cso-generation` manifest
|
||||
must be built and installed together for the current host. Windows also requires
|
||||
the empty `.gstack-cso-generation.lock` file from that same publication. Windows
|
||||
builds the native launcher with Visual Studio 2022 Build Tools (Desktop
|
||||
development with C++), discovered from Git Bash through the installed Developer
|
||||
PowerShell script; only the core is compiled with Bun.
|
||||
The dedicated Windows CI job checks startup, argument forwarding, and runtime
|
||||
injection separately from static audit and private-state support. POSIX
|
||||
watchdog execution remains unavailable there.
|
||||
|
||||
Every successful fresh install or upgrade asks the trusted launcher to preload
|
||||
the current host platform's qualified runtime and scanner images. This is the
|
||||
only automatic image acquisition path. It first validates the committed
|
||||
catalogs, pins a local Unix-socket Docker endpoint, rejects remote TCP/SSH
|
||||
contexts, and uses a new empty Docker configuration so pulls are anonymous.
|
||||
Only fully qualified `registry/repository@sha256:...` catalog entries are
|
||||
requested. Missing Docker or public registry access is a visible, nonfatal
|
||||
prerequisite: static audits remain available and a later `./setup` retries.
|
||||
The preload stage reserves 30 seconds for endpoint admission and gives each
|
||||
declared image a separate 30-second inspection/download window. Set
|
||||
`GSTACK_CSO_IMAGE_PULL_TIMEOUT_SECONDS=120` when registry speed requires a
|
||||
longer per-image window (accepted range: 5–300 seconds). The helper derives an aggregate deadline from the
|
||||
catalog size and caps the complete stage at one hour. A per-image timeout leaves
|
||||
that digest unavailable and continues with the next entry; reaching the hard
|
||||
aggregate deadline stops all remaining work. When a registry prerequisite stops
|
||||
further downloads, setup still checks later catalog entries locally and reports
|
||||
every already-present digest. A later setup resumes from exact local digests.
|
||||
Application, verifier, and scanner execution still use `--pull=never` with no
|
||||
egress. `gstack-cso doctor --repo PATH` never downloads; it reports a qualified
|
||||
profile unavailable until that exact digest and platform are already present
|
||||
in the local daemon.
|
||||
|
||||
The ordinary free suite exercises contracts without requiring Docker. The
|
||||
`cso-docker-integration` CI job sets `GSTACK_CSO_DOCKER_TESTS=1`; unavailable Docker
|
||||
or missing containment prerequisites fail that job. Run it locally with a
|
||||
working local daemon:
|
||||
|
||||
```bash
|
||||
bun run build:cso
|
||||
GSTACK_CSO_DOCKER_TESTS=1 bun run test:cso:docker
|
||||
```
|
||||
|
||||
When the staged-image workflow supplies `GSTACK_CSO_TEST_IMAGE`, platform,
|
||||
stack, and exact tool versions, the same command also runs stack-specific
|
||||
qualification journeys. A second native qualification job waits for every
|
||||
same-architecture staging job, so Rails receives both its application digest
|
||||
and the staged PostgreSQL digest; a missing matching artifact fails the gate.
|
||||
The Node journey starts with an empty public archive
|
||||
cache, acquires an integrity-pinned dependency through the registry-restricted
|
||||
broker, installs and executes it offline, reproduces a producer-visible fixture
|
||||
defect, and verifies a patch bound to its recorded closure and preparation
|
||||
proofs. The helper issues a `runtime_tested` `RepairBundle` only when its
|
||||
out-of-process witness authenticates the external assertions. Canonical project
|
||||
tests run in a target-controlled process, so their completion remains
|
||||
`self_reported`: target code can forge reporter output or terminate the runner.
|
||||
Recorded command, count, exit, and output hashes are diagnostic evidence; they
|
||||
do not establish target-independent test completion. The `tested` state remains
|
||||
reserved for a future target-independent witness and is not emitted today.
|
||||
Until the runtime-tested bundle, replay, and current-source recheck journey
|
||||
passes, the held-out repair gate remains pending and catalog
|
||||
promotion is prohibited. The candidate journey also asserts that no run-owned
|
||||
containers, preparation call directories, archive staging directories, or
|
||||
verification work directories remain. A skipped staged-image journey is not a
|
||||
passing cold-start or private held-out-repair result. Bun and Python have cold acquisition, offline install,
|
||||
boot, control, and project-test journeys. Rails runs those checks twice against
|
||||
the same cold public-gem inputs: once with two disposable SQLite connections
|
||||
and once with two databases in a fresh staged PostgreSQL sidecar. The Rails
|
||||
fixture builds the source-platform `sqlite3` and `pg` native gems only in the
|
||||
offline execution phase.
|
||||
|
||||
For hashed Python requirements resolved by pip, the broker records the requested
|
||||
index URL, allowed contacted hosts, and verified wheel hash. Pip's internal
|
||||
`files.pythonhosted.org` response URL is opaque to that TLS boundary, so
|
||||
`resolvedUrl` remains `null`; `registryResponseSha256` identifies the acquired
|
||||
archive bytes, not a verified response URL.
|
||||
|
||||
Image publication starts with the committed
|
||||
[`build-inputs.json`](../lib/cso/images/build-inputs.json). Its reviewed revision
|
||||
pins distinct native manifests for every stack and the SBOM generator. Each
|
||||
profile records the inspected source tag and immutable index for provenance,
|
||||
then `baseImages` keyed by `linux/amd64` and `linux/arm64`; Python additionally
|
||||
records native uv manifests. CI fails if a source tag no longer resolves to the
|
||||
reviewed index, a platform manifest is absent from that index, its image
|
||||
configuration names another architecture, or an executable reports a different
|
||||
runtime or package-manager version. Bun does not declare a Node version because
|
||||
its container's `node` path is a Bun-backed fallback rather than a Node release.
|
||||
A Bun project with a Node engine constraint therefore receives an explicit
|
||||
prerequisite instead of a fabricated compatibility result.
|
||||
|
||||
Configure the repository's `cso-runtime-release` GitHub environment with required
|
||||
maintainer review and deployments restricted to protected `main`. The
|
||||
[`cso-runtime-images.yml`](../.github/workflows/cso-runtime-images.yml) workflow
|
||||
runs a build-only native matrix on pull requests. Its manual publication jobs
|
||||
only accept `main`. They build on native amd64 and arm64 runners, publish to
|
||||
the `cso-staging` namespace, signs provenance and SBOM attestations, and verifies
|
||||
the workflow identity and source commit. It then pulls that exact digest into
|
||||
the local daemon for tests that prohibit implicit pulls. The per-image smoke test executes
|
||||
the embedded verifier against a loopback service with both a passing control
|
||||
and an intentionally failing control, alongside the containment suite. After
|
||||
all staging rows finish, native qualification jobs run the matching cold-start
|
||||
journey described above. The Node row additionally runs the complete lifecycle
|
||||
journey.
|
||||
Registry publication is staging, not runtime qualification.
|
||||
|
||||
GitHub Container Registry creates new packages private. Before a staging digest
|
||||
can enter native or private qualification, a package administrator must make
|
||||
its package public in GitHub's package settings. This is an explicit bootstrap
|
||||
step because GitHub documents the visibility change as irreversible and does
|
||||
not provide a supported workflow API for it. The release jobs use their scoped
|
||||
workflow token to require `visibility: public` from GitHub's package REST
|
||||
endpoint, then pull the exact platform digest with `docker --config` pointing
|
||||
at a new directory whose `config.json` contains only an empty `auths` object.
|
||||
The GitHub token, Docker credentials, contexts, certificate paths, and
|
||||
credential helpers are absent from that Docker client. A private package,
|
||||
metadata mismatch, failed anonymous pull, or different returned digest stops
|
||||
staging, qualification, and promotion. Both
|
||||
runtime and scanner promotion repeat this anonymous pull check on fresh hosted
|
||||
runners before opening a catalog PR.
|
||||
|
||||
Review the retained staging artifacts together with all checks in
|
||||
[`qualification.json`](../lib/cso/images/qualification.json). Cold-start fixtures,
|
||||
held-out repair evaluations, precision/recall thresholds, crash cleanup, and all
|
||||
secret-canary checks must pass before a reviewed catalog change can make a
|
||||
digest executable. A successful protected-main qualification run packages one
|
||||
`qualified-runtime.json` statement for each stack and architecture under the
|
||||
`cso-qualified-runtime-statements` artifact. All ten statements bind the same
|
||||
source commit and workflow run. The protected
|
||||
[`cso-runtime-promote.yml`](../.github/workflows/cso-runtime-promote.yml) workflow
|
||||
authenticates that run through the Actions API, checks every statement against
|
||||
the committed build profiles, and emits an attested
|
||||
`runtime-catalog.candidate.json`. The protected job immediately verifies the
|
||||
candidate attestation against its workflow identity, protected-main source ref,
|
||||
source commit, and exact file digest. It then compares `previousRevision` with
|
||||
the source catalog, commits those same bytes to a fresh branch, and opens a
|
||||
normal review PR. A concurrent catalog promotion fails the compare-and-swap or
|
||||
produces a merge conflict instead of silently replacing the newer matrix.
|
||||
The catalog stores a separately recomputable digest of the retained runtime
|
||||
matrix and the digest of the complete external qualification statements. The
|
||||
helper rejects a post-generation image or qualification mutation even if the
|
||||
statement digest field was left unchanged.
|
||||
Rollback retains the prior compatible catalog revision.
|
||||
|
||||
The public qualification ingress is
|
||||
[`cso-runtime-qualification.yml`](../.github/workflows/cso-runtime-qualification.yml).
|
||||
Set the protected environment variable `CSO_QUALIFICATION_ACTOR` to the GitHub
|
||||
service account used by the private evaluator. After its held-out assertions and
|
||||
accuracy run pass, that account sends a `repository_dispatch` event of type
|
||||
`cso-runtime-qualified`. The payload has one key, `statements`, containing ten
|
||||
objects with `schemaVersion`, `helperAbi`, `state`, `buildRevision`,
|
||||
`runtimeId`, `stack`, `platform`, immutable `image`, exact `versions`,
|
||||
`sourceCommit`, and the true-valued `checks` accepted by
|
||||
`cso-runtime-promotion.ts`. The ingress checks the actor and matrix size,
|
||||
re-verifies both OCI attestations against the protected staging workflow and
|
||||
source commit, replaces the workflow, time, SBOM digest, and provenance digest
|
||||
with values it observed, then runs the catalog generator as its final schema
|
||||
gate. It uploads the statements only after that gate passes. The payload carries
|
||||
no held-out assertion, application source, finding, or repair bundle.
|
||||
|
||||
The current reviewed profiles and unpromoted catalog establish this release
|
||||
process; they do not claim published gstack images, passed containment, or
|
||||
measured security accuracy.
|
||||
Private held-out find-to-repair evaluations remain mandatory release gates for
|
||||
Node, Bun, Python, and Rails. The producer-visible Node lifecycle fixture and the
|
||||
other cold-start journeys do not satisfy those gates. Qualification artifacts
|
||||
therefore record private checks as `"pending"` and `qualified: false` until all
|
||||
four private evaluations and the accuracy thresholds pass. The remaining
|
||||
external prerequisite is one successful protected-main run that uploads all ten
|
||||
authenticated statements. This sandbox cannot create that evidence because it
|
||||
has neither a native arm64 runner nor the private held-out assertions.
|
||||
|
||||
Attestation behavior follows [Docker's SBOM documentation](https://docs.docker.com/build/metadata/attestations/sbom/)
|
||||
and [GitHub CLI's verification policy](https://cli.github.com/manual/gh_attestation_verify).
|
||||
Public-package behavior follows [GitHub's package permissions documentation](https://docs.github.com/en/packages/learn-github-packages/about-permissions-for-github-packages),
|
||||
[GitHub's package REST endpoint](https://docs.github.com/en/rest/packages/packages#get-a-package-for-a-user),
|
||||
and [Docker's client configuration documentation](https://docs.docker.com/reference/cli/docker/),
|
||||
inspected on September 11, 2026.
|
||||
@@ -0,0 +1,103 @@
|
||||
# Running CSO scanners
|
||||
|
||||
`gstack-cso scan RUN SCANNER [REQUEST.json]` collects candidate evidence through
|
||||
the trusted helper when a matching qualified scanner catalog profile exists.
|
||||
Recognized IDs are `gitleaks`, `osv`, `semgrep`, `zizmor`, `trivy`, and
|
||||
`schemathesis`. Without a qualified profile, the command records a specific
|
||||
prerequisite; static assessment and importing existing CodeQL or other SARIF
|
||||
with `gstack-cso import-sarif RUN REPORT.sarif` remain available.
|
||||
The returned immutable artifact can be read later with
|
||||
`gstack-cso scanner-outcome RUN ARTIFACT_ID`; callers never need to open private
|
||||
state files directly.
|
||||
|
||||
The source-controlled catalog in
|
||||
[`scanner-images/catalog.json`](../lib/cso/scanner-images/catalog.json) is empty
|
||||
until every release gate produces real evidence. Without a reviewed profile,
|
||||
the command records a specific `not_assessed` prerequisite. Installing a
|
||||
scanner on the host does not enable execution. Each profile binds a Linux
|
||||
platform, immutable image digest, fixed entrypoint, absolute scanner executable,
|
||||
version-probe hash, required adapter capabilities, helper ABI, and the exact
|
||||
isolation-policy hash. The helper validates that catalog when it starts and has
|
||||
no executable fallback.
|
||||
Setup preloads the current platform's qualified scanner digests through the
|
||||
trusted installation helper using anonymous public-registry access. A failed
|
||||
preload or a bounded per-image/setup deadline is a nonfatal prerequisite and
|
||||
never causes an audit to pull. The default is 30 seconds per declared image;
|
||||
`GSTACK_CSO_IMAGE_PULL_TIMEOUT_SECONDS=120` selects a longer allowance (accepted
|
||||
range: 5–300 seconds) while the complete preload remains capped at one hour. Doctor is
|
||||
read-only and reports each scanner unavailable unless its exact catalog digest
|
||||
is already present locally.
|
||||
|
||||
All scanner processes use Docker containment with a watchdog, shared admission
|
||||
limits, read-only source, bounded output, and an isolated network namespace.
|
||||
Static scanners have no network access, including when the audit omits
|
||||
`--offline`. Semgrep rules and OSV/Trivy databases must be baked into the reviewed
|
||||
image. Their content hashes, scope, and freshness accompany the result. Database
|
||||
refresh is a separately reviewed image acquisition/release operation; scans
|
||||
never download missing assets or switch to unrestricted registry networking.
|
||||
|
||||
`REQUEST.json` can select an exact scanner `profile`. Schemathesis additionally
|
||||
requires comprehensive mode and an `api` object; `gstack-cso schema` describes
|
||||
its fields. The API harness supplies a bounded OpenAPI 3.0/3.1 JSON document,
|
||||
1–20 unique path operation IDs, a qualified application runtime, an absolute
|
||||
start command, a numeric loopback port, a legitimate HTTP control, and the
|
||||
source files defining the security boundary. Internal schema references are
|
||||
allowed. Server overrides, remote references/examples, hooks, callbacks, and
|
||||
webhooks are rejected. The application must boot and pass the control before
|
||||
the scanner starts. Application and scanner share only the admitted loopback
|
||||
namespace. An API failure is a candidate, never a reproduced vulnerability or
|
||||
certified repair. Sources needing dependency preparation report that prerequisite
|
||||
until a prepared offline closure is available to the API runner.
|
||||
|
||||
Each execution probes the scanner version before scanning and validates the
|
||||
retained source both before and after collection. Unknown output, failed
|
||||
diagnostics, a timeout, changed source, or incomplete cleanup cannot become an
|
||||
empty clean result. The helper retains valid partial candidates when the
|
||||
adapter can distinguish them from failed coverage. Output that cannot be
|
||||
redacted is withheld.
|
||||
|
||||
Scanner and SARIF collection hold the run's mutation lock and reject finished
|
||||
or interrupted reports. Every outcome is written exclusively under a unique
|
||||
`scanner-outcomes/` name before its coverage record enters the report. The
|
||||
returned `artifact` path is relative to the run directory. Model submissions
|
||||
cannot replace `scanner:*` coverage or certify the imported candidates. The
|
||||
free scanner tests use injected runners and catalogs to verify dispatch and
|
||||
failure handling; they do not claim real image qualification. The opt-in Docker
|
||||
qualification test runs the production adapter, image, watchdog, and isolation
|
||||
policy. It emits the version-output hash only after a representative scan
|
||||
completes without a coverage gap.
|
||||
|
||||
The release inputs in
|
||||
[`build-inputs.json`](../lib/cso/scanner-images/build-inputs.json) remain
|
||||
`pending` until reviewers provide all six scanners on both native platforms.
|
||||
The inputs require immutable base and SBOM-generator image digests, exact source
|
||||
commits and versions, signer workflow identities and digests, and reviewed
|
||||
canonical SLSA/SPDX statement-set digests. The workflow cryptographically
|
||||
re-verifies each image with those repository, source, signer, and predicate
|
||||
constraints before use. Semgrep additionally requires a reviewed local rules bundle. OSV and
|
||||
Trivy require complete offline database bundles with recorded content hashes,
|
||||
freshness, and ecosystem coverage. Publishing those asset-bearing base images
|
||||
and reviewing their evidence are external prerequisites; the release workflow
|
||||
does not invent or silently replace them.
|
||||
|
||||
[`cso-scanner-images.yml`](../.github/workflows/cso-scanner-images.yml) lets a
|
||||
dispatched branch run read-only input and contract validation. Image publishing,
|
||||
native amd64/arm64 qualification, attestation verification, embedded-asset hash
|
||||
checks, containment, real-adapter execution, and catalog-proposal generation run
|
||||
only from protected `main` through the `cso-scanner-release` environment. Before
|
||||
enabling qualification or promotion, configure the repository's
|
||||
`cso-scanner-release` GitHub environment with required maintainer reviewers and
|
||||
deployments restricted to protected `main`. GitHub otherwise creates a referenced
|
||||
missing environment without protection rules. Promotion requires an explicit
|
||||
dispatch from `main` and approval through that configured environment. It opens a pull request
|
||||
containing only the proposed source catalog, so code review remains the final
|
||||
authorization step. `previousRevision` records the prior compatible catalog for
|
||||
rollback and must still equal the current `main` catalog when promotion runs.
|
||||
|
||||
Scanner flags follow the primary documentation linked in
|
||||
[`scanners.ts`](../lib/cso/scanners.ts). Offline qualification must exercise
|
||||
the exact chosen version: OSV's maintainers documented a cache-location
|
||||
regression in 2.5.0, which demonstrates why a version string alone is insufficient.
|
||||
[OSV issue #2983](https://github.com/google/osv-scanner/issues/2983).
|
||||
Trivy uses an explicit memory scan cache alongside read-only baked databases.
|
||||
[Trivy filesystem reference](https://trivy.dev/docs/v0.68/guide/references/configuration/cli/trivy_filesystem/).
|
||||
+8
-8
@@ -23,7 +23,7 @@ Detailed guides for every gstack skill — philosophy, workflow, and examples.
|
||||
| [`/land-and-deploy`](#land-and-deploy) | **Release Engineer** | Merge the PR, wait for CI and deploy, verify production health. One command from "approved" to "verified in production." |
|
||||
| [`/canary`](#canary) | **SRE** | Post-deploy monitoring loop. Watches for console errors, performance regressions, and page failures in your Aside browser. |
|
||||
| [`/benchmark`](#benchmark) | **Performance Engineer** | Baseline page load times, Core Web Vitals, and resource sizes. Compare before/after on every PR. Track trends over time. |
|
||||
| [`/cso`](#cso) | **Chief Security Officer** | OWASP Top 10 + STRIDE threat modeling security audit. Scans for injection, auth, crypto, and access control issues. |
|
||||
| [`/cso`](#cso) | **Chief Security Officer** | Supported security findings with explicit coverage. Static assessment remains available without catalog profiles; contained runtime/scanner execution requires matching qualified profiles. Runtime-tested bundles authenticate separate external assertions. Project-test completion remains `self_reported` because target code controls the test process; `tested` is reserved for a future target-independent completion witness. |
|
||||
| [`/document-release`](#document-release) | **Technical Writer** | Update all project docs to match what you just shipped. Catches stale READMEs automatically. |
|
||||
| [`/document-generate`](#document-generate) | **Technical Writer** | Generate Diataxis docs (tutorial / how-to / reference / explanation) for a feature from code. |
|
||||
| [`/retro`](#retro) | **Eng Manager** | Team-aware weekly retro. Per-person breakdowns, shipping streaks, test health trends, growth opportunities. |
|
||||
@@ -760,19 +760,19 @@ Claude: Benchmarking 5 pages (3 runs each)...
|
||||
|
||||
This is my **Chief Security Officer**.
|
||||
|
||||
Run `/cso` on any codebase and it performs an OWASP Top 10 + STRIDE threat model audit. It scans for injection vulnerabilities, broken authentication, sensitive data exposure, XML external entities, broken access control, security misconfiguration, XSS, insecure deserialization, known-vulnerable components, and insufficient logging. Each finding includes severity, evidence, and a recommended fix.
|
||||
Run `/cso` for a bounded static investigation with an application model, challenged findings, and explicit coverage; static assessment remains available when no runtime or scanner catalog profile is qualified. With matching qualified profiles, `/cso --comprehensive` can prepare Node/Bun, Python, and Rails applications in contained local runtimes, reproduce a defect, and retain a reviewable repair candidate without changing the working branch. An out-of-process witness can authenticate the external boot, legitimate-control, and security assertions and issue a `runtime_tested` bundle. Project-test completion remains `self_reported` because target code shares that process and can forge reporter output or terminate the runner; recorded command, count, exit, and output hashes are diagnostic evidence, not a target-independent completion witness. Every report shows assertion, test-completion, and review assurance separately. The `tested` state is reserved for a future target-independent witness and is not emitted today. `/cso --doctor`, `--resume`, `--replay`, and `--recheck` diagnose prerequisites, recover interrupted work, repeat recorded verification, and establish current-source closure from fresh evidence.
|
||||
|
||||
```
|
||||
You: /cso
|
||||
|
||||
Claude: Running OWASP Top 10 + STRIDE security audit...
|
||||
Claude: complete — assessed application routes, tenant authorization, secrets,
|
||||
dependency exposure, and deployment configuration.
|
||||
|
||||
CRITICAL: SQL injection in user search (app/models/user.rb:47)
|
||||
HIGH: Session tokens stored in localStorage (app/frontend/auth.ts:12)
|
||||
MEDIUM: Missing rate limiting on /api/login endpoint
|
||||
LOW: X-Frame-Options header not set
|
||||
HIGH: Cross-tenant invoice access (app/controllers/invoices.rb:47)
|
||||
Confidence: high — caller, middleware, and policy checks traced
|
||||
Evidence: supported static finding; runtime not requested
|
||||
|
||||
4 findings across 12 files scanned. 1 critical, 1 high.
|
||||
1 supported finding. Run ID: cso-…
|
||||
```
|
||||
|
||||
---
|
||||
|
||||
Reference in New Issue
Block a user